Anthropic Chief Executive Officer Dario Amodei appeared on CBS Sunday Morning to answer criticism around so-called "AI doomer" arguments after a former employee disclosed safety concerns. Amodei said his team has continued to run defensive assessments on the possibility that rapidly advancing AI could be misused in biological terrorism, and he argued for independent review mechanisms as well as a more measured pace of development.
Amodei outlines steps to limit AI misuse
Amodei said there are several ways to reduce the risk that AI models are used improperly. One is to bring in third-party evaluation bodies to observe how models are trained and to check whether companies are following established safety rules. Another is to slow AI development to a degree so that defensive systems and human control can keep pace with growth.
He also said industry leaders need to work with governments to set safety standards. At the same time, he argued that the burden should not sit only with private companies and that governments should take part in supervision.
Calls for US-China talks similar to arms control
On international coordination, Amodei said geopolitical tensions make cooperation difficult, but he described it as a necessary AI defense strategy. He argued that the United States and China should enter negotiations similar to nuclear arms-control talks and build a shared safety framework.
He said preventing biological terrorism is in the interest of countries including the US and China. As one option, the two sides could first reach a consensus treaty or expand the Biological Weapons Convention framework to explicitly ban the use of AI in developing state biological weapons that threaten security.
Proposal to expand the Biological Weapons Convention
Amodei said that, under global competition and geopolitical pressure, countries should borrow from traditional arms-control models and create cross-border security communication channels. Using public health security as an example, he said countries share a common interest in preventing biological attacks. In that context, he suggested expanding the current Biological Weapons Convention, or BWC, to explicitly prohibit the use of AI models in designing dangerous pathogens and to make sure participating states follow through on safety commitments.
Key points highlighted in the report
- Independent third-party audits: bring in nonprofit organizations or government-backed outside evaluators to observe model training and operations.
- Adjusted development pace: slow the rate of progress to a degree so safety defenses and human oversight can keep up.
- Use existing arms-control experience: draw on nuclear and biological weapons control practice and extend the BWC approach into AI governance.
- Build real verification tools: rely on embedded inspection and technical monitoring to carry out and enforce cross-border agreements.
- Strengthen government-company coordination: combine government regulation, corporate self-discipline, and layered defenses to reduce the chance that one failed link triggers a wider disaster.
ABMedia also noted that the All in Podcast described this kind of doomer narrative as a psychological tactic and said it could affect Anthropic's listing plans.